06 - Raster manipulation and analysis
What you learn today? Wrangle your rasters
In these exercises, you learn how to work with raster data in R, specifically, how to
- review and recode problematic raster values
- crop raster extent and mask it to a boundary specified by a polygon
- reduce raster resolution
- extract values from raster by points, lines or polygons
- overlay multiple rasters and do raster algebra
- inspect prominence at locations and sort it
- sum raster values across vector shapes
Welcome to Bulgaria
In terms of data, today you will be working with an ASTER digital elevation model, an IKONOS multispectral satellite image and a prominence raster, all for the Kazanlak Valley in central Bulgaria. The images are provided by the JICA and GeoEye Foundations respectively, the prominence raster is derived from the elevation model:
- Aster image contains a digital elevation model for the area.
- IKONOS image is a 4-band satellite image covering ca 150 sq km of
the Kazanlak Valley with red, green, blue and infrared band, captured in
2001. As it is a fairly large orthophoto, I divided the original into
two halves with each half at 2 GB and placed them in ScienceData. If you
want to play with the full-sized images, you can download them manually
from public www.sciencedata.dk
folder, or directly with with
file.download()using these direct links for West and East respectively. You can also skip the large files and load a slightly reduced image of the East side of the ValleyKazE.tiffrom your data/ folder in the course of exercise 4. - prominence is a 30m resolution raster which assigns each cell a percentage that represents the number of cells in a radius of 1500m that are lower in elevation than the given cell.
Task 1: Access raster data values
Raster data can be very big depending on the extent and resolution
(grid size). In order to deal with this the raster() and
brick() functions are designed to only read in the actual
raster values as needed. To show that this is true, you can use the
inMemory() function on an object and it will return
FALSE if the values are not in memory. If you use the
head() function, the raster package will read
in only the values needed, not the full set of values. The raster values
will be read in by default if you perform spatial analysis operations
that require it or you can read in the values from a raster manually
with the function getValues().
Instructions
- Activate
rasterlibrary - Use
GDALinfo()to inspect the properties of the Aster image raster in the data folder (the file name isAster.tif). What can you learn from this inspection about its bands, resolution, and projection? - Read in the Aster image. Review Week 02 raster loading if unsure about how.
- Use the
inMemory()function on the elevation object to determine if the data has been read in. - Use the
head()function to look at the first few values from the elevation raster. - Use the
getValues()function on the elevation object to read in all the data. - Use the
hist()function to create a quick histogram of the elevation values. Note the pile of values near -9999, these should beNA(any idea why?) and we will address this later.
# library
___(raster)
# Read in the elevation layer
elevation <- ___("data/Aster.tif")
# Check if the data is in memory
___(___)
# Use head() to peak at the first few records
___(___)
# Use getValues() to read the values into a vector
vals <- ___(elevation)
# Use hist() to create a histogram of the values
___(vals)Solution
[1] FALSE
1 2 3 4 5 6 7 8 9 10 11 12
1 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
2 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
3 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
13 14 15 16 17 18 19 20
1 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
2 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
3 -9999 -9999 -9999 -9999 -9999 -9999 -9999 -9999
[ reached getOption("max.print") -- omitted 7 rows ]
Congratulations! You now know that the raster package
only reads in raster values as needed to save space in memory. You can
get the values manually using the getValues() function.
Now, a new question arises, why are so many values encoded as -9999? Are we in the Mariana Trench all of sudden?
Task 2: Change values: handle missing or bad data values in rasters
There are many situations where you might need to recode raster
values. You may want to change the outlier values to NA for
example. In the raster package, reclassification is
performed with the reclassify() function.
In the elevation raster you’ve worked with the values
are meters above sea level and are supposed to range between 0 and 2500.
Anything below 0 should be an NA. In this exercise you will
assign any values below 0 to NA.
Instructions
- Check that the package
rasterand the objectelevationare loaded in your workspace. - Plot the
elevationraster usingplot(). - Set up a three-column matrix with the
cbind()function and values -10000, 0, NA. - Use the matrix and
reclassify()to reclassify values below 0 toNA. You will need to use the argumentrcl. - Plot the reclassified elevation layer to confirm there are no values below 0.
# Plot the elevation layer to see the legend values
___(elevation)
# Set up the matrix
vals <- ___
# Reclassify
elevation_reclass <- ___(elevation, ___ = ___)
# Plot again and confirm that the legend range is 0 - 2400
___(elevation_reclass)Solution
Good work! Knowing how to reclassify rasters will come in handy. When
you get a chance you should review the help for
reclassify() particularly the part that discusses how to
specify the rcl argument. The three-column approach from
this exercise is most common but there are other approaches.
Task 3: Crop and mask rasters on the basis of other spatial objects
Mask and crop are two similar operations that allow you to limit your
raster to a specific area of interest. With crop() you
limit the extent of your raster to that of your focus area. With
mask() you essentially place your area of interest on top
of the raster and any raster cells outside of the boundary are assigned
NA values.
The Aster image covers a large area, but we are primarily interested in the areas surveyed by archaeologists, the first of which is the cluster of burial mound points and second, large survey polygons registered by local archaeologists during field survey. In this exercise you will use the extent of mounds and survey units to crop and mask the elevation raster.
On some OSs, raster package does not support
sf objects. This should not happen now, but if you
encounter difficulty with raster:vector interactions, it helps to know
that you can convert the vector to Spatial object with, for
example, as(input, "Spatial").
Instructions I - Crop raster by the extent of the mounds dataset
- Load
sflibrary - Create
moundsobject from the shapefile “KAZ_mounds.shp”. For a refresher on vector loading, check out Week 02 instructions. - Verify that
moundsCRS matches theelevation_reclassCRS and fix withst_transform()if not. - Create a bounding box
mounds_bbaround the mounds withst_make_grid()following Week 03 guidelines. - Plot the
mounds_bband themoundsto see how these two objects relate. - Crop the
elevation_reclasslayer by themoundsobject using thecrop()function and create a smallerelevobject - Plot the
elevandmoundsand themounds_bbtogether.
Solution
[1] "Modes: S4, character"
[2] "Attributes: < Modes: list, NULL >"
[3] "Attributes: < Lengths: 3, 0 >"
[4] "Attributes: < names for target but not for current >"
[5] "Attributes: < current is not list-like >"
Instructions II - Filter largest survey polygons
- Create
surveyobject from a shapefile called “KAZ_units.shp” - Project the
surveyobject to match the newelevraster withst_transform()if needed. - Compute the area of the survey with
st_area()and save this object asareas. What units are these? - Filter the survey units to only those above 30 ha with the
filter()function. You will need to wrapareasinunclass(). Save assurvey_big. Remember to have the tidyverse or dplyr library attached forfilter()to work properly. Alsosfmight interfere so specify thedplyr::filter()if needed.
# Read in the survey object
survey <- ___(___)
# Compute the area of the survey
areas <- ___(survey)
# Filter to survey with areas > 30000
survey_big <- ___(survey, ________ > 30000)Solution
Instructions III - Mask raster by the largest polygons
- Review the plot of
elevraster. - Plot the geometry of the
survey_bigover it. - Mask the
elevlayer withsurvey_bigand save aselevation_mask. This may take a couple of seconds. - Review the plot of
elevation_mask.
# Plot the elevation raster
plot(________)
# Plot the geometry of survey_big
plot(_________(survey_big))
# Mask the elev layer with survey_big and save as elevation_mask
elevation_mask <- ________(elev, mask = survey_big)
# Plot the elevation_mask -- this is a raster!
plot(elevation_mask)Solution
# Mask the elev layer with survey_big and save as elevation_mask
elevation_mask <- mask(elev, mask = survey_big)
# Plot the elevation_mask -- this is a raster!
plot(elevation_mask)Nice! You ensured that layers had the same CRS, you cropped the
raster, computed bounding boxes and areas for masking and filtered the
data. Finally, you used mask() to mask the elevation raster
to show only the large survey units.
Question:
1. What extent does the elevation raster default to after cropping by mounds vs by masking by the large polygons of survey_big?
2. Why do we not use the mounds bounding box to crop the elevation raster?
Task 4: Reduce the raster resolution (grid cell size) using aggregate()
Rasters, such as orthophotos, terrain models, or mosaiced rasters for large area, often come in resolutions far greater than you need (browse examples athttps://datafordeler.dk/dataoversigt/danmarks-hoejdemodel-dhm/overflade-praedefineret-geotiff/) . Up till now you have played with fairly small, processed rasters. Now you are getting a taste of the real thing. To reduce the computational load when running analyses, or when developing the right approach, you should use a reduced resolution raster.
The function to reduce resolution in rasters is
aggregate() which, as you might guess, aggregates grid
cells into larger grid cells using a user-defined function (for example,
mean or max). The function used to aggregate the values is determined by
the fun argument (the default being mean) and the amount of
aggregation is driven by the fact argument (the default
being 2).
Instructions
- Load the East half of the two IKONOS images of the Kazanlak Valley
from
data/KazE.tif. See Welcome to Bulgaria for links to full-resolution rasters. - Plot the file you read in with the appropriate function for a multi-band raster.
- Determine the raster resolution using
res()and number of raster cells in the layer withncell(). - Aggregate the IKONOS image using the default for
funand with afactorof 10 and save the new raster toeast_small. - Plot the new raster for comparison to the old version.
- Determine the new raster resolution and the number of raster cells.
# Load the IKONOS raster
east <- ________________
# Plot the IKONOS raster
___(east)
# Determine the raster resolution
___(east)
# Determine the number of cells
___(east)
# Aggregate the raster
east_small <- ___(east, fact = ___)
# Plot the new east layer
___(east_small)
# Determine the new raster resolution
___(east_small)
# Determine the number of cells in the new raster
___(east_small)Solution
# Load the IKONOS raster
east <- brick("../data/KazE.tif")
# Plot the IKONOS raster
plotRGB(east, stretch = "lin")[1] 10 10
[1] 2374596
# Aggregate the raster
east_small <- aggregate(east, fact = 10)
# Plot the new east layer
plotRGB(east_small, stretch = "lin")[1] 100 100
[1] 23940
Lovely job! In this example you read in a raster and then converted it to a lower resolution raster to save on the size of the object and ultimately computation power required. In this example, the raster was not too big to begin with so perhaps aggregating would not be necessary but for big rasters such as you will see in remote sensing session (next week), this can be a big help and a necessity.
Task 5: Extract raster values by location
Beyond simply masking and cropping you may want to know the actual
cell values at locations of interest. You might, for example, want to
know the elevation at each mound location or perhaps the mean elevation
within the large survey units. This is where the extract()
function comes in handy.
Usefully, and you’ll see this in a later analysis, you can feed
extract() a function that will get applied to extracted
cells. For example, you can use extract() to extract raster
values by point, line, polygon, or neighborhood and with the
fun = mean argument it will return an average cell value by
neighborhood.
Instructions
- Ensure
moundsandelevorelevationobjects are still in memory. - Use the
rasterfunctionextract()to determine the elevation at each of the mounds. Assign the extracted value into aelevcolumn in the mounds object. Beware that theextract()function exists across multiple packages, so it’s wise to useraster::in front of it. - Look at the
moundselevation column through theplot()function as well as histogram. Do theextract()results make sense? Theelevationlayer values represent meters above sea level.
# Extract the elevation values at the mound locations
________$elev <- _______::______(elevation, mounds)
# Look at the mounds and extraction results
____(mounds[_______])
hist(____)Solution
Great! raster::extract() is a very useful tool. It can
be used with polygons as well as points and lines, and the result can be
written out as a separate vector or added to an existing object as a
column.
Task 6: Raster math with overlay
You will now use the elevation layer and a “prominence”
layer. Prominence measures what percentage of surrounding cells are
below any given location in the raster. A high percentage value thus
indicates a prominent point, while a low percentage indicates a
low-lying location with poor inter-visibility. Archaeologists assert
that mounds have been built in elevated places with a good field of
view, so let’s see whether they are mostly right :)
What you will do in this exercise is essentially: identify the most
prominent locations among the registered archaeological mounds by
finding areas that have both a high percentage of prominence and a high
elevation. You will use two rasters and you will define an overlay
function f to do the raster math with. Remember that raster
algebra uses basic Boolean logic to select desirable values. Note that
you can only do raster math on two rasters of the same extent,
so make sure you align the elevation and prominence accordingly with the
crop() function
Instructions
- Make sure the
elevationobject exists in memory (“Aster.tif”, it is a single-band raster). - Read in the prominence raster layer from “prominence1500.tif”; it is also a single-band raster.
- Plot the
prominenceobject. Do the legend units make sense? - Specify function
f, where you select elevation values greater than 400 msl and below 650m (mounds don’t appear above that elevation) and prominence values over 60% - Call
overlay()onelevationandprominence. Set thefunargument tof. - If you have not cropped the two rasters to the same extent yet, you
can do so inside the
overlay()function
# Check in on the elevation and read in prominence layer
elevation
prominence <- ___(___)
# Plot prominence
plot(prominence)
# Function f with 2 arguments and the raster math to select specific elevation range and prominence values
f <- function(rast1, rast2) {
rast1 _________ & rast2 ___________
}
# Align the extent of the two rasters with crop()
__________
# Do the overlay using the above define f function in the `fun` argument.
prom_el_overlay <- ___(______, ___, fun = ___)
# Plot the result (low elevation and high prominence areas)
___(prom_el_overlay)Solution
class : RasterLayer
dimensions : 590, 627, 369930 (nrow, ncol, ncell)
resolution : 30, 30 (x, y)
extent : 352483.7, 371293.7, 4712336, 4730036 (xmin, xmax, ymin, ymax)
crs : +proj=utm +zone=35 +datum=WGS84 +units=m +no_defs
source : memory
names : Aster
values : 265, 1302 (min, max)
Congratulations! You’ve now learned to perform raster math using the
raster function overlay(). You limited to areas with
>400m & <650m elevation and >60% prominence, these areas
should harbor the most prominent mounds (or defense-ready areas if
associated with settlements) in Kazanlak.
Questions:
3. What is the actual value range in the
prom_el_overlay raster?
4. What area is covered by each cell?
Task 7: Inspect the most prominent mounds
In the above exercise, we produced an elevation-prominence overlay. Mounds and other sites that sit in this overlay enjoy a strategic position vis-a-vis the rest of the valley. Which ones are they, however, and what are their real prominence values? It is hard to see at the scale of the valley and it would be good to pull the sites out.
Find the mounds that enjoy the most prominent locations as well as those that feature in the elevation-prominence overlay raster. Produce a list of the ID numbers (TRAP_Code) of all the overlay mounds and 25 most prominent mounds and plot them (expressing their prominence somehow) .
Instructions
Check that you have the
moundssf object,prominenceandprom_el_overlayrasters.Plot the
prom-el-overlayand themoundson top of each other to check visual overlap. Do the same withprominenceraster andmounds.Extract values from the elevation-prominence overlay raster and from the prominence raster at mound locations and write them to two columns:
– call the first column
prom_el_overlay– name the second columnprominenceMake an object of mounds that sit within these strategic high-visibility locations. How many are there?
Make an object of 25 mounds with the highest prominence values (remember
arrange()andslice()?). WhichTRAP_Codeids are included?Plot both these sets of mounds using the
mapview()library and compare their locations.
Solution
Question:
5. How do the mounds with high prom_el_overlay values differ from those with high prominence?
Task 8: Practice further on Aarhus municipality
Load the Danish elevation file. Crop the elevation by Aarhus
municipality boundary or by local forests. Calculate a summary of
elevation among the urban shelters (shelters.rds)